CHAPTER 7. Introduction to Intuitionistic and Modal Logics. 1 Introduction to Intuitionictic Logic

Transcription

1 CHAPTER 7 ch7 Introduction to Intuitionistic and Modal Logics 1 Introduction to Intuitionictic Logic Intuitionistic logic has developed as a result of certain philosophical views on the foundation of mathematics, known as intuitionism. Intuitionism was originated by L. E. J. Brouwer in The first Hilbert style formalization of the intuitionistic logic, formulated as a proof system, is due to A. Heyting (1930). In this chapter we present a Hilbert style proof system I that is equivalent to the Heyting s original formalization and discuss the relationship between intuitionistic and classical logic. There have been, of course, several successful attempts at creating semantics for the intuitionistic logic,i.e. to define formally a notion of the intuitionistic tautology. The most recent are Kripke models were defined by Kripke in The first intuitionistic semantics was defined in a form of pseudo- Boolean algebras by McKinsey, Tarski in McKinsey, Tarski algebraic approach to the intuitionostic semantics (and classical) was followed by many authors and developed into a new field of Algebraic Logic. The pseudo- Boolean algebras are called also Heyting algebras. An uniform presentation of algebraic models for classical, intuitionistic and modal logics was first given in a now classic algebraic logic book: Mathematics of Metamathematics, Rasiowa, Sikorski (1964). The goal of this chapter is to give a presentation of the intuitionistic logic formulated as a proof system, discuss its algebraic semantics and the basic theorems that establish the relationship between classical and intuitionistic logics. 1.1 Philosophical Motivation Intuitionists view-point on the meaning of the basic logical and set theoretical concepts used in mathematics is different from that of most mathematicians in their research. The basic difference lies in the interpretation of the word exists. For example, let A(x) be a statement in the arithmetic of natural numbers. For the mathematicians the sentence xa(x) (1) s 1

2 is true if it is a theorem of arithmetic, i.e. if it can be deduced from the axioms of arithmetic by means of classical logic. If a mathematician proves sentence ( 1), this does not mean that he is able to indicate a method of construction of a natural number n such that A(n) holds. For the intuitionist the sentence (1) is true only he is able to provide a constructive method of finding a number n such that A(n) is true. Moreover, the mathematician often obtains the proof of the existential sentence (1), i.e. of the sentence xa(x) by proving first a sentence x A(x). (2) s1 Next he makes use of a classical tautology ( x A(x)) xa(x)). (3) s2 By applying Modus Ponens to (2) and (3) he obtains (1). For the intuitionist such method is not acceptable, for it does not give any method of constructing a number n such that A(n) holds. For this reason the intuitionist do not accept the classical tautology (3) i.e. ( x A(x)) xa(x)) as intuitionistic tautology, or as as an intuitionistically provable sentence. Let us denote by I A and = I A the fact that A is intuitionistically provable and that A is intuitionistic tautology, respectively. The proof system I for the intuitionistic logic has hence to be such that I ( x A(x)) xa(x)). The intuitionistic semantics I has to be such that one can prove in that also = I ( x A(x)) xa(x)). The above means also that intuitionists interpret differently the meaning of propositional connectives. Intuitionistic implication The intuitionistic implication (A B) is considered by to be true if there exists a method by which a proof of B can be deduced from the proof of A. In the case of the implication ( x A(x)) xa(x)) there is no general method which, from a proof of the sentence ( x A(x)), permits is to obtain an intuitionistic proof of the sentence xa(x), i.e. to construct a number n such that A(n) holds, hence we can t accept it as an intuitionistic theorem or tautology. 2

3 Intuitionistic negation The negation and the disjunction are also understood differently. The sentence A is considered intuitionistically true if the acceptance of the sentence A leads to absurdity. As a result of above understanding of negation and implication we have that in the intuitionistic logic I I (A A) but I ( A A). Consequently, the intuitionistic semantics I has to be such that = I (A A) and = I ( A A). Intuitionistic disjunction The intuitionist regards a disjunction (A B) as true if one of the sentences A, B is true and there is a method by which it is possible to find out which of them is true. As a consequence a classical law of excluded middle (A A) is not acceptable by the intuitionists since there is no general method of finding out, for any given sentence A, whether A or A is true. This means that the intuitionistic logic must be such that I (A A) and the intuitionistic semantics I has to be such that = I (A A). Intuitionists view of the concept of infinite set also differs from that which is generally accepted in mathematics. Intuitionists reject the idea of infinite set as a closed whole. They look upon an infinite set as something which is constantly in a state of formation. Thus, for example, the set of all natural numbers is infinite in the sense that to any given finite set of natural numbers it is always possible to add one more natural number. The notion of the set of all subsets of the set of all natural numbers is not regarded meaningful. Thus intuitionists reject the general idea of a set as defined by a modern set theory. An exact exposition of the basic ideas of intuitionism is outside the range of our investigations. Our goal is to give a presentation of of the intuitionistic logic, which is a sort of reflection of intuitionistic ideas formulated as a proof system. 3

4 1.2 Algebraic Intuitionistic Semantics and Completeness Theorem There are many proof systems describing the intuitionistic logic. We define now a system I with a set of axioms that is due to Rasiowa (1959). We adopted this axiomatization for two reasons. Firs is that it is the most natural and appropriate set of axioms to carry the the algebraic proof of the completeness theorem and the second is that they visibly describe the main difference between intuitionistic and classical logic. Namely, by adding the only one more axiom (A A) we get a (complete) formalization for classical logic. Here are the components if the proof system I. Language We adopt a propositional language L = L {,,, } with the set of formulas denoted by F. Axioms A1 A2 A3 A4 A5 A6 A7 A8 A9 A10 A11 ((A B) ((B C) (A C))), (A (A B)), (B (A B)), ((A C) ((B C) ((A B) C))), ((A B) A), ((A B) B), ((C A) ((C B) (C (A B))), ((A (B C)) ((A B) C)), (((A B) C) (A (B C)), (A A) B), ((A (A A)) A), where A, B, C are any formulas in L. Rules of inference We adopt the Modus Ponens rule (MP ) A ; (A B) B as the only inference rule. 4

5 A proof system I = ( L, F A1 A11, (MP ) ), (4) Isys A1 A11 defined above, is called a Hilbert style formalization for Intuitionistic propositional logic. We introduce, as usual, the notion of a formal proof in I and denote by I A the fact that A has a formal proof in I, or that that A is provable in I. intuitionistically 1.3 Algebraic Semantics and Completeness Theorem We shortly present here Tarski, Rasiowa, and Sikorski psedo-boolean algebra semantics and discuss the algebraic completeness theorem for the intuitionistic propositional logic. We leave the Kripke semantics for the reader to explore from other, multiple sources. Here are some basic definitions. Here are some basic definitions. Relatively Pseudo-Complemented Lattice A lattice (B,, ) is said to be relatively pseudo-complemented (Birkhoff, 1935) if for any elements a, b B, there exists the greatest element c, such that a c b. Such element is denoted by a b and called the pseudo-complement of a relative to b. By definition x a b if and only if a x b for all x, a, b B. (5) p-c The equation (5) can serve as the definition of the relative pseudo-complement a b. unit1 Fact 1 Every relatively pseudo-complemented lattice (B,, ) has the greatest element, called a unit element and denoted by 1. Proof Observe that a x a for all x, a B. By (5) we have that x a a for all x B, i.e. a a = 1. An abstract algebra B = (B, 1,,,, ) (6) rpc is said to be a relatively pseudo-complemented lattice if (B,, ) is relatively pseudo-complemented lattice with the relative pseudo-complement defined by (5) and with the unit element 1 (Fact 1 ). 5

6 Relatively Pseudo-complemented Set Lattices Consider a topological space X with an interior operation I. Let G(X) be the class of all open subsets of X and G (X) be the class of all both dense and open subsets of X. Then the algebras (G(X), X,,, ), (G (X), X,,, ), where, are set-theoretical operations of union, intersection, and is defined by Y Z = I(X Y ) Z are relatively pseudo-complemented lattices. Clearly, all sub algebras of these algebras are also relatively pseudo-complemented lattices, called relatively pseudo-complemented set lattices. They are typical examples of relatively pseudo-complemented lattices Pseudo - Boolean Algebra (Heyting Algebra) An algebra B = (B, 1, 0,,,, ), (7) Pb-alg is said to be a pseudo - Boolean algebra if and only if (B, 1,,, ) it is a relatively pseudo-complemented lattice (6) in which a zero element 0 exists and is a one argument operation defined as follows a = a 0 (8) i-neg The operation is called a pseudo-complementation. The pseudo - Boolean algebras are also called Heyting algebras to stress their connection to the intuitionistic logic. Let X be topological space with an interior operation I. Let G(X) be the class of all open subsets of X. Then (G(X), X,,,,, ), (9) p-f where, are set-theoretical operations of union, intersection, is defined by and is defined as Y Z = I(X Y ) Z Y = Y = I(X Y ), for all Y X is a pseudo - Boolean algebra. Every sub algebra of G(X) is also a pseudo-boolean algebra. They are called pseudo-fields of sets. 6

7 The following Theorem 1 states that pseudo-fields are typical examples of pseudo - Boolean algebras. The theorems of this type are often called Stone Representation Theorems to remember an American mathematician H.M Stone. He was one of the first to initiate the investigations between logic and general topology in the article The Theory of Representations for Boolean Algebras,Trans. of the Amer.Math, Soc 40, thm:repr Theorem 1 (Representation Theorem) (McKinsey, Tarski, 1946) For every pseudo - Boolean (Heyting) algebra (7) B = (B, 1, 0,,,, ), there exists a monomorphism h of B into a pseudo-field (9) G(X) of all open subsets of a compact topological T 0 space X. Another typical (and interesting) example of a class of pseudo - Boolean algebras is the following. Linear Pseudo - Boolean Algebra Let (B, ) be a chain (linearly ordered set) with the greatest element 1 and the least element (smallest) 0. An algebra H = (B, 1, 0,,,, ) (10) Lpb is called a linear pseudo - Boolean algebra if and only if its operations are defined as follows. For any a, b B, a b = max{a, b}, a b = min{a, b}, { 1 if a b a b = b otherwise, (11) L-impl and define the pseudo-complementation as a = a 0. We leave the proof that (10) is a pseudo-boolean algebra as a homework exercise. Observe that the linear pseudo - Boolean algebra (10) is a generalization of the 3-valued Heyting semantics defined in chapter??. Algebraic Models We say that a formula A is an intuitionistoc tautology if and only if any pseudo-boolean algebra (7) is a model for A. This kind of models because their connection to abstract algebras are called algebraic models. We put it formally as follows. 7

8 I-model Definition 1 (Intuitionistic Algebraic Model) Let A be a formula of the language L {,,, } and let B = (B, 1, 0,,,, ) be a pseudo - Boolean topological Boolean algebra (7). We say that the algebra B is a model for the formula A and denote it by B = A if and only if v (A) = 1 holds for all variables assignments v : V AR B. I-taut Definition 2 (Intuitionistic Tautology) The formula A is an intuitionistic tautology and is denoted by for all pseudo-boolean algebras B. = I A if and only if B = A In Algebraic Logic the notion of tautology is often defined using a notion a formula A is valid in an algebra B. It is formally defined as follows. I-valid Definition 3 A formula A is valid in a pseudo-boolean algebra B = (B, 1, 0,,,, ), if and only if v (A) = 1 holds for all variables assignments v : V AR B. Directly from definition 2 and definition 3 we get the following. Ivalid Fact 2 For any formula A, = I A if and only if A is valid in all pseudo- Boolean algebras. We write now I A to denote any proof system for the Intuitionistic propositional logic, and in particular the Hilbert style formalization for Intuitionistic propositional logic I defined by (4). I-com Theorem 2 (Intuitionistic Completeness Theorem) (Mostowski 1948) For any formula A of L {,,, }, I A if and only if = I A. The intuitionistic completeness theorem 2 follows also directly from the following general algebraic completeness theorem 3 that combines results of of Mostowski (1958), Rasiowa (1951) and Rasiowa-Sikorski (1957). 8

9 I-thm Theorem 3 (Algebraic Completeness Theorem) For any formula A of L= L {,,, } the following conditions are equivalent. (i) I A, (ii) = I A, (iii) A is valid in every pseudo-boolean algebra (G(X), X,,,,, ) of open subsets of any topological space X, (iv) A is valid in every pseudo-boolean algebra B with at most 2 2r elements, where r is the number of all sub formulas of A. Moreover, each of the conditions (i) - (iv) is equivalent to the following one. (v) A is valid in the pseudo-boolean algebra (G(X), X,,,,, ) of open subsets of a dense-in -itself metric space X (in particular of an n- dimensional Euclidean space X). The following theorem follows from the equivalence of conditions (i) and (iv). Idec Theorem 4 (Decidability) Every proof system for the intuitionistic propositional logic is decidable. Examples of intuitionistic propositional tautologies The following classical tautologies are provable in I and hence are also intuitionistic tautologies. (A A), (12) i1 (A (B A)), (13) i2 (A (B (A B))), (14) i3 ((A (B C)) ((A B) (A C))), (15) i4 (A A), (16) i5 (A A), (17) i6 (( A B) (A B)), (18) i7 ( (A B) ( A B)), (19) i8 (( A B) ( (A B)), (20) i9 (( A B) ( A B)), (21) i10 ((A B) ( B A)), (22) i11 9

10 ((A B) (B A)), (23) i12 ( A A), (24) i13 ( A A), (25) i14 ( (A B) (A B)), (26) i15 ((C A) ((C (A B)) (C B)), (27) i16 Examples of classical tautologies that are not intuitionistic tautologies The following classical tautologies are not intuitionistic tautologies. (A A), (28) ni1 ( A A), (29) ni2 ((A B) ( A B)), (30) ni3 ( (A B) ( A B)), (31) ni4 (( A B) ( B A)), (32) ni5 (( A B) (B A)), (33) ni6 ((A B) A) A), (34) ni7 1.4 Connection Between Classical and Intuitionistic Tautologies The intuitionistic logic has been created as a rival to the classical one. So a question about the relationship between these two is a natural one. We present here some examples of tautologies and some historic results about the connection between the classical and intuitionistic logic. The first connection is quite obvious. It was proved by Rasiowa and Sikorski in 1964 that by adding the axiom A12 (A A) to the set of axioms of our system I defined by (4) we obtain a Hilbert proof system H that is complete with respect to classical semantics. This proves the following. 10

11 it1 Theorem 5 Every formula that is derivable intuitionistically is classically derivable, i.e. if I A, then A, where we use symbol for classical (complete classical proof system) provability. We write = A and ] = I A to denote that A is a classical and intuitionistic tautology, respectively. As both proof systems, I and H are complete under respective semantics, we can state this as the following relationship between classical and intuitionistic tautologies. it2 Theorem 6 For any formula A F, if = I A, then = A. The next relationship shows how to obtain intuitionistic tautologies from the classical tautologies and vice versa. It has been proved by Glivenko in 1929 and independently by Tarski in it3 Theorem 7 (Glivenko, Tarski) For any formula A F, A is a classically provable if and only if A is an intuitionistically provable, i.e. I A if and only if A where we use symbol for classical provability. it4 Theorem 8 (McKinsey, Tarski, 1946) For any formula A F, A is a classical tautology if and only if A is an intuitionistic tautology, i.e. = A if and only if = I A. The following relationships were proved by Gödel. 11

12 it5 Theorem 9 (Gödel, 1931) For any A, B F, a formula (A B) is a classically provable if and only if it is an intuitionistically provable, i.e. (A B) if and only if I (A B). it6 Theorem 10 (Gödel, 1931) If a formula A contains no connectives except and, then A is a classically provable if and only if it is an intuitionistically provable. By the completeness of classical and intuitionisctic logics we get the following equivalent semantic form of theorems 9 and 10. it7 Theorem 11 A formula (A B) is a classical tautology if and only if it is an intuitionistic tautology, i.e. = (A B) if and only if = I (A B). it8 Theorem 12 If a formula A contains no connectives except and, then A is a classical tautology if and only if it is an intuitionistic tautology. On intuitionistically derivable disjunction In a classical logic it is possible for the disjunction (A B) to be a tautology when neither A nor B is a tautology. The tautology (A A) is the simplest example. This does not hold for the intuitionistic logic. This fact was stated without the proof by Gödel in 1931 and proved by Gentzen in 1935 via his proof system LI which is presented in chapter?? and discussed in the next section 2. it9 Theorem 13 (Gentzen 1935) A disjunction (A B) is intuitionistically provable if and only if either A or B is intuitionistically provable, i.e. I (A B) if and only if I A or I B. 12

13 We obtain, via the Completeness Theorem 2 the following equivalent semantic version of the above. it10 Theorem 14 A disjunction (A B) is intuitionistic tautology if and only if either A or B is intuitionistic tautology, i.e. = I (A B) if and only if = I A or = I B. G-LI 2 Gentzen Sequent System LI In 1935 G. Gentzen formulated a first syntactically decidable formalizations for classical and intuitionistic logic and proved its equivalence with the Heyting s original Hilbert style formalization. He named his classical system LK (K for Klassisch) and intuitionistic system LI (I for Intuitionistisch). In order to prove the completeness of the system LK and proving the adequacy of LI he introduced a special rule, called cut rule that corresponds to the Modus Ponens rule in Hilbert proof systems. Then, as the next step he proved the now famous Gentzen Hauptzatz, called in English the Cut Elimination Theorem. The Gentzen original proof system LI is a particular case of his proof system LK for the classical logic. Both of them were presented in chapter?? together with the proof of the Hauptzatz for both, LK and LI systems. The elimination of the cut rule and the structure of other rules makes it possible to define effective automatic procedures for proof search, what is impossible in a case of the Hilbert style systems. The Gentzen system LI is defined as follows. Language of LI Let SQ = { Γ : Γ, F } be the set of all Gentzen sequents built out of the formulas of the language L = L {,,, } (35) I-lan and the additional symbol. In order to describe the the intuitionistic logic we deal, after Gentzen, only with sequents of the form Γ, where consists of at most one formula. I.e. we assume that all LI sequents are elements of a following subset ISQ of the set SQ of all sequents. ISQ = {Γ : consists of at most one formula }. (36) iseq The set ISQ is called the set of all intuitionistic sequents; the LI sequents. 13

14 Axioms of LI As the axioms of LI we adopt any sequent from the set ISQ defined by ( 36), which contains a formula that appears on both sides of the sequent arrow, i.e any sequent of the form Γ 1, A, Γ 2 A, (37) iax for any formula A F of the language (35) and for any sequences Γ 1, Γ 2 F. Inference rules of LI The set inference rules is divided into two groups: the structural rules and the logical rules. They are defined as follows. Structural Rules of LI Weakening ( weak) Γ Γ A. A is called the weakening formula. Contraction (contr ) A, A, Γ A, Γ, A is called the contraction formula, contains at most one formula. Exchange (exchange ) Γ 1, A, B, Γ 2 Γ 1, B, A, Γ 2, contains at most one formula. Logical Rules of LI Conjunction rules ( ) A, B, Γ (A B), Γ, ( ) Γ A ; Γ B, Γ (A B) contains at most one formula. Disjunction rules ( ) 1 Γ A Γ (A B), ( ) 2 Γ B Γ (A B), 14

15 ( ) A, Γ ; B, Γ, (A B), Γ contains at most one formula. Implication rules ( ) A, Γ B Γ (A B), ( ) Γ A ; B, Γ (A B), Γ, contains at most one formula. Negation rules ( ) Γ A A, Γ, ( ) A, Γ Γ A. Formally we define: LI = (L, ISQ, LA, Structural rules, Logical rules ), (38) LI where ISQ is defined by (36), Structural rules and Logical rules are the inference rules defined above, and LA is the axiom defined by the schema (37). We write LI Γ to denote that the sequent Γ has a proof in LI. We say that a formula A F has a proof in LI and write it as LI A when the sequent A has a proof in LI, i.e. LI A if and only if LI A. The completeness of of our cut-free LI follows directly from LI Hauptzatz Theorem proved in chapter?? and the Intuitionistic Completeness Theorem 2. The proof is a straightforward adaptation of the proof of cut free LK Completeness Theorem proved in chapter?? and is left as a homework exercise. LI-compl Theorem 15 (Completeness of LI) For any sequent Γ ISQ, LI Γ if and only of = I Γ. 15

16 In particular, for any formula A, LI A if and only of = I A. G-disj Theorem 16 (Intuitionistically Derivable Disjunction) For any formulas A, B, LI (A B) if and only if LI A or LI B. In particular, a disjunction (A B) is intuitionistically provable in any proof system I if and only if either A or B is intuitionistically provable in I. The particular form the theorem 16 was stated without the proof by Gödel in The theorem proved by Gentzen in 1935 via his Hauptzatz Theorem. Proof Assume LI (A B). This equivalent to LI (A B). The last step in the proof of (A B)i LI must be the application of the rule ( ) 1 to the sequent A, or the application of the rule ( ) 2 to the sequent B. There is no other possibilities. We have proved that LI (A B) implies LI A or LI B. The inverse is obvious by respective applications of rules ( ) 1 ( ) 2 to A and B. sec:dtrees 2.1 Decomposition Trees in LI Search for proofs in LI is a much more complicated process then the one in classical systems RS or GL defined in chapter??. Here, as in any other Gentzen style proof system, proof search procedure consists of building the decomposition trees. In RS the decomposition tree T A of any formula A, and hence of any sequence Γ is always unique. In GL the blind search defines, for any formula A a finite number of decomposition trees, but it can be proved that the search can be reduced to examining only one of them, due to the absence of structural rules. In LI the structural rules play a vital role in the proof construction and hence, in the proof search. We consider here a number of examples to show the complexity of the problem of examining possible decomposition trees for a given formula A. We are going to see that the fact that a given decomposition tree ends with an axiom leaf does not always imply that the proof does not exist. It might only imply that our search strategy was not good. Hence the problem of deciding whether a given formula A does, or does not have a proof in LI becomes more complex then in the case of Gentzen system for classical logic. 16

17 Before we define a heuristic method of searching for proof and deciding whether such a proof exists or not in LI we make some observations. Observation 1: the logical rules of LI are similar to those in Gentzen type classical formalizations we examined in previous chapters in a sense that each of them introduces a logical connective. Observation 2: The process of searching for a proof is, as before a decomposition process in which we use the inverse of logical and structural rules as decomposition rules. For example the implication rule: ( ) A, Γ B Γ (A B) becomes an implication decomposition rule (we use the same name ( ) in both cases) ( ) Γ (A B) A, Γ B. Observation 3: we write our proofs in as trees, instead of sequences of expressions, so the proof search process is a process of building a decomposition tree. To facilitate the process we write, as before, the decomposition rules, structural rules included in a tree form. For example the the above implication decomposition rule is written as follows. Γ (A B) ( ) A, Γ B The two premisses implication rule ( ) written as the tree decomposition rule becomes (A B), Γ ( ) Γ A B, Γ 17

18 For example the structural weakening rule is written as the decomposition rule is written as ( weak) Γ A Γ We write it in a tree form as follows. Γ A ( weak) Γ We define, as before the notion of decomposable and indecomposable formulas and sequents as follows. Decomposable formula is any formula of the degree 1. Decomposable sequent is any sequent that contains a decomposable formula. Indecomposable formula is any formula of the degree 0, i.e. any propositional variable. Remark: In a case of formulas written with use of capital letters A, B, C,.. etc, we treat these letters as propositional variables, i.e. as indecomposable formulas. Indecomposable sequent is a sequent formed from indecomposable formulas only. Decomposition tree construction (1): given a formula A we construct its decomposition tree T A as follows. Root of the tree is the sequent A. Given a node n of the tree we identify a decomposition rule applicable at this node and write its premisses as the leaves of the node n. We stop the decomposition process when we obtain an axiom or all leaves of the tree are indecomposable. Observation 4: the decomposition tree T A obtained by the construction (1) most often is not unique. 18

19 Observation 5: the fact that we find a decomposition tree T A with non-axiom leaf does not mean that LI A. This is due to the role of structural rules in LI and will be discussed later in the chapter. We illustrate the problems arising with proof search procedures, i.e. decomposition trees construction in the next section 2.2 and give a heuristic proof searching procedure in the section 2.3. sec: examles 2.2 Proof Search Examples We perform proof search and decide the existence of proofs in LI for a given formula A F by constructing its decomposition trees T A. We examine here some examples to show the complexity of the problem. Remark In the following and similar examples when building the decomposition trees for formulas representing general schemas we treat the capital letters A, B, C, D... as propositional variables, i.e. as indecomposable formulas. Example 1 Determine whether LI (( A B) (A B)). This means that we have to construct some, or all decomposition trees of (( A B) (A B)). If we find a decomposition tree such that all its leaves are axioms, we have a proof. If all possible decomposition trees have a non-axiom leaf, proof of A in LI does not exist. Consider the following decomposition tree of (( A B) (A B)). T1 (( A B) ( (A B)) ( ) ( A B) (A B) ( ) (A B), ( A B) (exch ) ( A B), (A B) ( ) 19

20 A, B, (A B) ( ) B, (A B) A ( weak) B, (A B) ( ) (A B) B ( ) A B non axiom B B axiom The tree T1 has a non-axiom leaf, so it does not constitute a proof in LI. But this fact does not yet prove that proof doesn t exist, as the decomposition tree in LI is not always unique. Let s consider now the following tree. T2 (( A B) ( (A B)) ( ) ( A B) (A B) ( ) (A B), ( A B) (exch ) ( A B), (A B) ( ) A, B, (A B) (exch ) A, (A B), B (exch ) (A B), A, B ( ) 20

21 A, A, B (exch ) A, A, B ( ) A, B A axiom B, A, B (exch ) B, B, A (exch ) B, B, A ( ) B, A B axiom All leaves of T2 are axioms, what proves that T2 is a proof of A and hence we proved that LI (( A B) (A B)). Example 2 Part 1: Prove that LI (A A), Part 2: Prove that LI ( A A). Solution of Part 1 To prove that LI (A A) we have to construct some, or all decomposition trees of (A A). We treat the sub formulas A, B as indecomposable formulas. Consider the following decomposition tree. T (A A). ( ) A A ( ) A, A 21

22 ( ) A A axiom All leaves of T are axioms what proves that T is a proof of (A A) and we don t need to construct other decomposition trees. Solution of Part 2 To prove that LI ( A A) we have to construct all decomposition trees of (A A) and show that each of them has an non-axiom leaf. Consider the first decomposition tree defined as follows. T1 ( A A) first of 2 choices : ( ), ( weak) ( ) A A first of 2 choices : ( weak), (contr ) ( weak) A first of 2 choices : ( ), (contr ) ( ) A first of 2 choices : ( ), ( weak) ( ) A indecomposable non axiom We use the first tree created to define all other possible decomposition trees by exploring the alternative search paths as indicated at the nodes of the tree. 22

23 T1 ( A A) ( ) one of 2 choices A A ] (contr ) second of 2 choices A, A A ( weak) first of 2 choices A, A ( ) first of 2 choices A A ( ) the only choice A, A (exch ) the only choice A, A ( ) the only choice A A ( ) first of 2 choices A, A indecomposable non axiom 23

24 We can see from the above decomposition trees that the blind construction of all possible trees only leads to more complicated trees, due to the presence of structural rules. Observe that the blind application of (contr ) gives an infinite number of decomposition trees. To decide that none of them will produce a proof we need some extra knowledge about patterns of their construction, or just simply about the number useful of application of structural rules within the proofs. In this case we can just make an external observation that the our first tree T1 is in a sense a minimal one; that all other trees would only complicate this one in an inessential way, i.e. we will never produce a tree with all axioms leaves. One can formulate a deterministic procedure giving a finite number of trees, but the proof of its correctness require some extra knowledge. We are going to discuss a motivation and an heuristics for the proof search in the next section. Within the scope of this book we accept the external explanation for the heuristics we use as a sufficient solution. As we can see from the above examples structural rules and especially the (contr ) rule complicates the proof searching task. The Gentzen type proof systems RS and GL from chapter don t contain the structural rules and are complete with respect to classical semantics, as is the original Gentzen system LK, which does contain the structural rules. As (via Completeness Theorem) all three classical proof system RS, GL, LK are equivalent we can say that the structural rules can be eliminated from the system LK. A natural question of elimination of structural rules from the intutionistic Gentzen system LI arizes. The following example illustrates the negative answer. Example 3 We know, by the theorem about the connection between classical and intuitionistic logic (theorem 6) and corresponding Completeness Theorems that for any formula A F, = A if and only if I A, where = A means that A is a classical tautology, I means that A is intutionistically provable, i.e. is provable in any intuitionistically complete proof system. The system LI is intuitionistically complete, so we have that for any formula A, = A if and only if LI A. We have just proved that LI ( A A). Obviously = ( A A), so we know that ( A A) must have a proof in LI. 24

25 We are going to prove that LI ( A A) and that the structural rule (contr ) is essential to the existence of its proof, i.e. that without it the formula ( A A) is not provable in LI. The following decomposition tree T is a proof of ( A A) in LI. T ( A A) first of 2 choices : ( ), ( weak) ( ) ( A A) first of 2 choices : (contr ), ( ) (contr ) ( A A), ( A A) one of 2 choices ( ) ( A A) ( A A) one of 3 choices ( ) ( A A), A A one of 2 choices ( weak) ( A A), A one of 3 choices (exch ) A, ( A A) one of 3 choices ( ) ( A A) A one of 3 choices ( ) 25

26 A, ( A A) one of 2 choices (exch ) ( A A), A one of 3 choices ( ) A ( A A) one of 3 choices ( ) A, A A axiom Assume now that the rule (contr ) is not available. All possible decomposition trees are as follows. T1 ( A A) ( ) one of 2 choices ( A A) ( ) only one choice ( A A) ( ) one of 2 choices A A ( weak) only one choice A ( ) only one choice A ( ) one of 2 choices A non axiom 26

27 T2 ( A A) ( weak) second of 2 choices non axiom T3 ( A A) ( ) ( A A) ( ) ( A A) ( weak) second of 2 choices non axiom T4 ( A A) ( ) ( A A) ( ) ( A A) ( ) ] A A ( weak) only one choice A ( ) 27

28 only one choice A ( weak) second of 2 choices non axiom This proves that the formula ( A A) is not provable in LI without (contr ) rule and hence this rule can t be eliminated. sec:hproc 2.3 Proof Search Heuristic Method Before we define a heuristic method of searching for proof in LI let s make some additional observations to the observations 1-5 from section 2.1. Observation 6: Our goal while constructing the decomposition tree is to obtain axiom or indecomposable leaves. With respect to this goal the use logical decomposition rules has a priority over the use of the structural rules and we use this information while describing the proof search heuristic. Observation 7: all logical decomposition rules ( ), where denotes any connective, must have a formula we want to decompose as the first formula at the decomposition node, i.e. if we want to decompose a formula A, the node must have a form A, Γ. Sometimes it is necessary to decompose a formula within the sequence Γ first in order to find a proof. For example, consider two nodes and n 1 = A, (A B) B n 2 = (A B), A B. We are going to see that the results of decomposing n 1 and n 2 differ dramatically. Let s decompose the node n 1. Observe that the only way to be able to decompose the formula A is to use the rule ( weak) first. The two possible decomposition trees that starts at the node n 1 are as follows. T1 n1 A, (A B) B 28

29 ( weak) A, (A B) ( ) (A B) A ( ) A, B A ( ) A, A, B non axiom T2 n1 A, (A B) B ( weak) A, (A B) ( ) (A B) A ( ) A, (A B) ( ) A, A, B non axiom Let s now decompose the node n 2. Observe that following our Observation 6 we start by decomposing the formula (A B) by the use of the rule ( ) first. A decomposition tree that starts at the node n 2 is as follows. T1 n2 (A B), A B ( ) A, B, A B axiom 29

30 This proves that the node n 2 is provable in LI, i.e. LI (A B), A B. Of course, we have also that the node n 1 is also provable in LI, as one can obtain the node n 2 from it by the use of the rule (exch ). Observation 8: the use of structural rules are important and necessary while we search for proofs. Nevertheless we have to use them on the must basis and set up some guidelines and priorities for their use. For example, use of weakening rule discharges the weakening formula, and hence an information that may be essential to the proof. We should use it only when it is absolutely necessary for the next decomposition steps. Hence, the use of weakening rule ( weak) can, and should be restricted to the cases when it leads to possibility of the use of the negation rule ( ). This was the case of the decomposition tree T1 n1. We used it as an necessary step, but still it discharged too much information and we didn t get a proof, when proof of the node existed. In this case the first rule in our search should have been the exchange rule, followed by the conjunction rule (no information discharge) not the weakening (discharge of information) followed by negation rule. The full proof of the node n 1 is the following. T3 n1 A, (A B) B (exch ) (A B), A B ( ) A, B, A B axiom As a result of the observations 1-5 from section 2.1 and observations 6-8 above we adopt the following. Heuristic Procedure for Proof Search in LI. For any A F we construct the set of decomposition trees T A following the rules below. 30

31 Rules for Decomposition Tree Generation 1. Use first logical rules where applicable. 2. Use (exch ) rule to decompose, via logical rules, as many formulas on the left side of as possible. 3. Use ( weak) only on a must basis in connection with ( ) rule. 4. Use (contr ) rule as the last recourse and only to formulas that contain or as connectives. 5. Let s call a formula A to which we apply (contr ) rule a contraction formula. 6. The only contraction formulas are formulas containing between theirs logical connectives. 7. Within the process of construction of all possible trees use (contr ) rule only to contraction formulas. 8. Let C be a contraction formula appearing on the node n of the decomposition tree of T A. For any contraction formula C, any node n, we apply (contr ) rule the the formula C at most as many times as the number of sub-formulas of C. If we find a tree with all axiom leaves we have a proof, i.e. LI A and if all (finite number) trees have a non-axiom leaf we have proved that proof of A does not exist, i.e. LI A. 3 Introduction to Modal S4 and S5 Logics The non-classical logics can be divided in two groups: those that rival classical logic and those which extend it. The Lukasiewicz, Kleene, and Intuitionistic Logics are in the first group. The modal logics are in the second. The rival logics do not differ from classical logic in terms of the language employed. Rather, rival logics differ in that certain theorems or tautologies of classical logic are rendered false, or not provable in them. Perhaps the most notorious example of this is the law of excluded middle (A A). This is provable in, and is a tautology of classical logic but is not provable in, and is not tautology of intuitionistic logic, or is not a tautology under any of the extensional logics semantics we have discussed. Logics which extend classical logic sanction all the theorems of classical logic but, generally, supplement it in two ways. Firstly, the languages of these nonclassical logics are extensions of those of classical logic, and secondly, the theorems of these non-classical logics supplement those of classical logic. Usually, 31

32 such supplementation is provided by the enriched language. For example, modal logics are enriched by the addition of two new connectives that represent the meaning of it is necessary that and it is possible that. We use the notation I for it is necessary that and C for it is possible that. Other notations used are:, N, L for it is necessary that, and P, M for it is possible that. The symbols N, L, P, M or alike, are often used in computer science investigations. The symbols and were first to be used in modal logic literature, the symbols I, C come from algebraic and topological interpretation of modal logics. I corresponds to the interior of the set and C to its closure. The idea of a modal logic was first formulated by an American philosopher, C.I. Lewis in He has proposed yet another interpretation of lasting consequences, of the logical implication. In an attempt to avoid, what some felt, the paradoxes of semantics for classical implication which accepts as true that a false sentence implies any sentence he created a notion of a modal truth, which lead to the notion of modal logic. The idea was to distinguish two sorts of truth: necessary truth and mere possible (contingent) truth. A possibly true sentence is one which, though true, could be false. A necessary truth is hence the one which could not be otherwise; a contingent (possible) truth is one which could. The distinction between them is a metaphysical one and should not be confused with the distinction between a priori and a posteriori truths. An a priori truth is one which can be known independently of experience, and an a posteriori truth is one which cannot. Such notions appeal to epistemic considerations and the whole area of modal logics bristles with philosophical difficulties and hence the numbers of logics have been created. Unlike the classical connectives, the modal connectives do not admit of truth-functional interpretation. This was the reason for which modal logics was first developed as a proof systems, with intuitive notion of semantics expressed by the set of adopted axioms. The first semantics, and hence the proofs of the completeness theorems came some 20 years later. It took yet another 25 years for discovery and development of the second more general approach to the semantics. These are two established ways of interpret modal connectives, i.e. to define modal semantics. The historically first one is due to Mc Kinsey and Tarski (1944, 1946). It is a topological interpretation that provides a powerful mathematical interpretation of some of them, namely S4 and S5. It connects the modal notion of necessity with the topological notion of interior of a set, and the notion of possibility with the notion of its closure. Our choice of symbols I and C for modal connectives comes from this interpretation. The topological interpretation powerful as it is, is less universal in providing models for other modal logics. The most recent one is due to Kripke (1964). It uses the notion possible world. Roughly, we say that CA is true if A is true in some possible world, called actual world, and IA is true if A is true in every possible world. We present the formal definition later in this chapter, but this intuitive meaning can be useful in unconvincing ourselves about validity (or sense) of adopted 32

33 axioms and rules of inference. As we have already mentioned, modal logics were first developed, as was the intuitionistic logic, in a form of proof systems only. First Hilbert style modal proof system was published by Lewis and Langford in They presented a formalization for two modal logics, which they called S1 and S2. They also outlined three other proof systems, called S3, S4, and S5. In 1933 Gödel worked with Heyting s sentential logic proof system, what we are calling now Intuitionistic logic. He considered a particular modal proof system and asserted that theorems of Heyting s sentential logic could be obtained from it by using a certain translation. His presentation of the discovered proof system, now known as S4 logic, was particularly elegant. Since then hundreds of modal logics have been created. There are some standard texts in the subject. These are, between the others: Hughes and Cresswell (1969) for philosophical motivation for various modal logics and Intuitionistic logic, Bowen (1979) for a detailed and uniform study of Kripke models for modal logics, Segeberg (1971) for excellent classification, and Fitting (1983), for extended and uniform studies of automated proof methods for classes of modal logics. Hilbert Style Modal Proof Systems We present here Hilbert style formalization for S4 and S5 logics due to Mc Kinsey and Tarski (1948) and Rasiowa and Sikorski (1964). We also discuss the relationship between S4 and S5, and between the Intuitionistic logic and S4 modal logic, as first observed by Gödel. They stress the connection between S4, S5 and topological spaces which constitute models for them. Hence the use of symbols I, C for necessity and possibility, respectively. The connective I corresponds to the symbol denoting a topological interior of a set and C to the closure of a set. Modal Language We add to the propositional language L {,,, } two extra one argument connectives I and C. I.e. we adopt L = L {,,,,I,C} (39) mlang as our modal language. We read a formula IA, CA as necessary A and possible A, respectively. The language is common to all modal logics. Modal logics differ on a choice of axioms and rules of inference, when studied as proof systems and on a choice of semantics. McKinsey, Tarski (1948) As modal logics extend the classical logic, any modal logic contains two groups 33

34 of axioms: classical and modal. Axioms Group 1: classical axioms Any modal logic adopts as its classical axioms any complete set of axioms for a classical propositional logic. Axioms Group 2: modal axioms M1 M2 M3 M4 (IA A), (I(A B) (IA IB)), (IA IIA), (CA ICA). Rules of inference We adopt the Modus Ponens (MP ) (MP ) A ; (A B) B and an additional modal rule (I) introduced by Gödel referred to as necessitation. (I) A IA We define modal proof systems S4 and S5 as follows. S4 = ( L, F, classical axioms, M1 M3, (MP ), (I) ), (40) s4 S5 = ( L, F, classical axioms, M1 M4, (MP ), (I) ). (41) s5 Observe that the axioms of S5 extend the axioms of S4 and both system share the same inference rules, hence we have immediately the following. mfact Fact 3 For any formula A F, if S4 A, then S5 A. Rasiowa, Sikorski (1964) It is often the case, and it is in our systems, that modal connectives are expressible by each other, i.e. that we can adopt one of them and define the other as follows. IA = C A, (42) mi and CA = I A. (43) mc 34

35 The equality sign in equations (42), and (43) means that we replace the formula in left side of the equation by the formula in the right side anywhere where the left side (or right side) formula is appears as a sub formula of a formula of L. In modal logics S4 and S5 the connective C is expressible by and I, as stated above by (43), we hence assume now that the language L contains only one modal connective I. Language L = L {,,,,I}. (44) l1 There are, as before, two groups of axioms: classical and modal. Axioms Group 1: classical axioms We adopt as classical axioms any complete set of axioms for a classical propositional logic. Axioms Group2: modal axioms R1 R2 R3 R4 R5 rfact Fact 4 ((IA IB) I(A B)), (IA A), (IA IIA), I(A A), ( I A I I A) Rules of inference We adopt the Modus Ponens (MP ) (MP ) and an additional modal rule ( RI) (RI) A ; (A B) B (A B) (IA IB). We define modal logic proof systems RS4, RS5 as follows. RS4 = ( L, F, classical axioms, R1 R4, (MP ), (RI) ). (45) rs4 RS5 = ( L, F, classical axioms, R1 R5, (MP ), (RI) ). (46) rs5 Observe that the axioms of RS5 extend, as the axioms of TS5, the axioms of TS4 and both system share the same inference rules, hence we have immediately the following. For any formula A F, if RS4 A, then RS5 A. 35

36 3.1 Algebraic Semantics for S4 and S5 The McKinsey, Tarski proof systems (40), (41), and Rasiowa, Sikorski proof systems (45), (45) for modal logics S4, S5 are complete with the respect to both algebraic topological semantics, and Kripke semantics. We shortly discuss the topological semantics, and algebraic completeness theorems and leave the Kripke semantics for the reader to explore from other, multiple sources. The topological semantics was initiated by McKinsey, Tarski s (1946, 1948) and consequently developed and examined by many authors into a field of Algebraic Logic.They are presented in detail in now classic algebraic logic books: Mathematics of Metamathematics, Rasiowa, Sikorski (1964) and An Algebraic Approach to Non-Classical Logics, Rasiowa (1974). We want to point out that the first idea of a connection between modal propositional calculus and topology is due to Tang Tsao -Chen, (1938) and Dugunji (1940). Here are some basic definitions. Boolean Algebra An abstract algebra B = (B, 1, 0,,,, ) (47) Balg is said to be a Boolean algebra if it is a distributive lattice and every element a B has a complement a B. Topological Boolean algebra By a topological Boolean algebra we mean an abstract algebra B = (B, 1, 0,,,,, I), (48) Talg where (B, 1, 0,,,, ) is a Boolean algebra (47) and, moreover, the following conditions hold for any a, b B. I(a b) = Ia Ib, Ia a = Ia, IIa = Ia, and I1 = 1. (49) Icond The element Ia is called a interior of a. The element I a is called a closure of a and will be denoted by Ca. Thus the operations I and C are such that Ca = I a and Ia = C a. In this case we write the topological Boolean algebra (48) as B = (B, 1, 0,,,,, I, C). (50) Talg1 It is easy to prove that in in any topological Boolean algebra (50) the following conditions hold for any a, b B. 36

37 C(a b) = Ca Cb, Ca a = Ca, CCa = Ca, and C0 = 0. (51) Ccond If X is a topological space with an interior operation I, then the family P(X) of all subsets of X is a topological Boolean algebra with 1 = X, the operation defined by the formula Y Z = (X Y ) Z, for all subsets Y, Zof X, set-theoretical operations of union, intersection, complementation, and the interior operation I. Obviously, every sub algebra of this algebra is a topological Boolean algebra, called a topological field of sets or, more precisely, a topological field of subsets of X. Given a topological Boolean algebra (47) (B, 1, 0,,,, ). The element a B is said to be open (closed) if a = Ia (a = Ca). Clopen Topological Boolean algebra A topological Boolean algebra (50) B = (B, 1, 0,,,,, I, C ). such that every open element is closed and every closed element is open, i.e. such that for any a B is called a clopen topological Boolean algebra. CIa = Ia and ICa = Ca (52) clopen We loosely say that a formula A of a modal language is a modal S4 tautology if and only if any topological Boolean algebra (50) is a model for A. We say that A is a modal S5 tautology if and only if any clopen topological Boolean algebra (52) is a model for A. We put it formally as follows. Amodel Definition 4 (Modal Algebraic Model) For any formula A of a modal language L {,,,,I,C} and for any topological Boolean algebra B = (B, 1, 0,,,,, I, C), the algebra B is a model for the formula A and denote it by B = A if and only if v (A) = 1 holds for all variables assignments v : V AR B. Mtaut Definition 5 (S4, S5 Tautology) The formula A is a modal S4 tautology (S5 tautology) and is denoted by = S4 A ( = S5 A) 37

38 if and only if for all topological Boolean algebras (clopen topological Boolean algebras) B we have that B = A (B = A). In Algebraic Logic the notion of tautology is often defined using a notion a formula A is valid in an algebra B.It is formally defined in our case as as follows. valid Definition 6 A formula A is valid in a topological Boolean algebra B = (B, 1, 0,,,,, I, C), if and only if v (A) = 1 holds for all variables assignments v : V AR B. Directly form definitions 5, 6 we get the following. Fact 5 For any formula A, = S4 A ( = S5 A) if and only if A is valid in all topological Boolean algebras (A is valid in all clopen topological Boolean algebras). We write S4 A and S5 A do denote any proof system for modal S4, S5 logics and in particular the proof systems (40), (45), and (41), (46), respectively. cs4 Theorem 17 (Completeness Theorem) For any formula A of the modal language L {,,,,I,C}, S4 A if and only if = S4 A, S5 A if and only if = S5 A. The completeness for S4 follows directly from the Theorem 18. The completeness for S5 follows from the S4 completeness and Embedding Theorems 22, 23. It also can be easily proved independently by adopting the Algebraic Completeness Theorem proof for S4 to clopen topological algebras. cthm Theorem 18 (Algebraic Completeness Theorem) For any formula A of the modal language L= L {,,,,I,C} the following conditions are equivalent. (i) S4 A, (ii) = S4 A, (iii) A is valid in every topological field of sets B(X), 38